Merge pull request #406 from diaoling6665/main

允许修改openai接口的超参

main/xiaozhi-server/config.yaml

@@ -142,6 +142,7 @@ VAD:
     min_silence_duration_ms: 700  # 如果说话停顿比较长，可以把这个值设置大一些
 
 LLM:
+  # 所有openai类型均可以修改超参，以AliLLM为例
   # 当前支持的type为openai、dify、ollama，可自行适配
   AliLLM:
     # 定义LLM API类型
@@ -150,6 +151,11 @@ LLM:
     base_url: https://dashscope.aliyuncs.com/compatible-mode/v1
     model_name: qwen-turbo
     api_key: 你的deepseek web key
+    temperature: 0.7  # 温度值
+    max_tokens: 500   # 最大生成token数
+    top_p: 1         
+    top_k: 50        
+    frequency_penalty: 0  # 频率惩罚
   DoubaoLLM:
     # 定义LLM API类型
     type: openai

main/xiaozhi-server/core/providers/llm/openai/openai.py

@@ -11,6 +11,12 @@ class LLMProvider(LLMProviderBase):
             self.base_url = config.get("base_url")
         else:
             self.base_url = config.get("url")
+        self.top_p = config.get("top_p")
+        self.top_k = config.get("top_k")
+        self.temperature = config.get("temperature")
+        self.max_tokens = config.get("max_tokens")
+        self.frequency_penalty = config.get("frequency_penalty")
+        
         check_model_key("LLM", self.api_key)
         self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)
 
@@ -19,7 +25,12 @@ class LLMProvider(LLMProviderBase):
             responses = self.client.chat.completions.create(
                 model=self.model_name,
                 messages=dialogue,
-                stream=True
+                stream=True,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens,
+                top_p=self.top_p,
+                top_k=self.top_k,
+                frequency_penalty=self.frequency_penalty
             )
             
             is_active = True
@@ -51,6 +62,11 @@ class LLMProvider(LLMProviderBase):
                 messages=dialogue,
                 stream=True,
                 tools=functions,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens,
+                top_p=self.top_p,
+                top_k=self.top_k,
+                frequency_penalty=self.frequency_penalty
             )
             
             for chunk in stream:
@@ -58,4 +74,4 @@ class LLMProvider(LLMProviderBase):
                 
         except Exception as e:
             self.logger.bind(tag=TAG).error(f"Error in function call streaming: {e}")
-            yield {"type": "content", "content": f"【OpenAI服务响应异常: {e}】"}
+            yield {"type": "content", "content": f"【OpenAI服务响应异常: {e}】"}